Nvidia's capacity commitments argue against AI orders drying up soon. Its filings show mostly memory and manufacturing commitments rising from $119 billion to $279 billion in one quarter, with about $175 billion covering 2027-28. Fewer new data-center projects could reflect power, land and building bottlenecks rather than weaker demand; a mid-teens multiple still reflects a fair worry that profits may not last. x.com 

Power-semiconductor content per AI server rack is set to rise about 17-fold, from $11,000 at B200 to $191,000 at Feynman, and the market has not seriously priced it in. As racks pass 200kW and power shifts from 48V to 800V DC, every conversion stage needs more power devices. VRMs and PSUs make up nearly 70% of a Feynman rack's content, Infineon's core turf, where its DrMOS and smart power stages are hard to displace. Design-in leads shipments by 12-18 months, so Rubin Ultra orders start now. … Show more

Infineon, STMicroelectronics and NXP have little value upside left from humanoid robotics after rallying hard on the AI datacenter theme. Infineon's CEO said humanoids could eventually rival AI datacenters as a revenue opportunity, about 20% of sales, yet Infineon saw insider sales this week. Humanoids may create few real bottlenecks: about 60% of parts come from the existing auto supply chain, most value will likely accrue in China and OEMs will integrate vertically, leaving heavy competition. … Show more

Semiconductors could stay undersupplied for a long time as more capable AI per dollar creates more uses and drives compute demand. Micron and NVIDIA trade at low forward P/Es because investors fear peak cyclical earnings, yet current earnings show no slowdown. Shortages may worsen as AI adoption spreads, though the industry could remain cyclical; a reversal of the forced rollover trade is hoped for, not assured. x.com  x.com  x.com 

SK hynix's contracted capacity gives its AI-memory demand case more visibility. It reported an 80% operating margin on ₩104.6 trillion of revenue, with sales more than quadrupling year over year and over half of 2027 and 2028 capacity under long-term contracts. Vertiv's sales grew 61%, orders more than doubled, and it raised full-year guidance again; so far, AI demand is still accelerating. x.com 

AI data-center investment needs outside AI spending to rise 33-fold by 2030 to break even. Hyperscalers hold $4.5 trillion in debt and future liabilities, but corporate and SaaS token spending outside circular deals is only $60 billion annualized against a $2 trillion break-even ARR. SK hynix says memory shortages may last until 2031, raising capex and CoreWeave default risk; Mitsubishi warns turbines are booked through 2031 and cost 400% more, delaying CoreWeave and Oracle data centers. x.com  … Show more

Nvidia retains pricing power if its full system keeps reducing customers' cost per effective AI task. Strong AI demand alone does not ensure profits across the chain: rising chip, memory and optical costs burden buyers, while tighter financing and slow collections threaten operators. Large customers developing their own chips could pressure Nvidia's margins even as the market grows. x.com 

AI suppliers' demand may not justify today's investment. Roughly $3 trillion in revenue absent last year is needed to support the spending, yet GDP growth is weak and many companies may be seeing only incremental gains. If the estimate that Anthropic uses two-thirds of its compute for training is right, it could triple inference supply. The most likely outcome for many companies, roughly half the S&P 500, is slowing growth, compressed margins and multiples; when the bubble ends is uncertain. … Show more

AI suppliers' profits and signed demand distinguish this buildout from dotcom overcapacity. AI leaders rose 2.9x on 3.1x earnings growth versus 7.4x on 2.1x before 2000; cloud providers have $2.3T in signed backlog. Alphabet, Microsoft, Amazon and Meta could repay debt with about six months' cash flow, unlike 2008's leveraged banks. Micron has multiyear take-or-pay deals and ASML has orders years ahead. Memory profits are cyclical; TSMC, ASML and GE Vernova remain favored in AI's early innings. … Show more

More AI tokens mean more compute demand for Nvidia even as pricing power starts to slip. OpenRouter data via the FT show weekly token use up ~250-fold since January 2025; the FT links the surge to agents' long call chains. Bloomberg's breakdown of a 1 GW AI campus that could cost roughly $30B puts well over half the bill in chips and servers, most landing at Nvidia as GPU racks. Servers wear out in 5-6 years, making AI capex a recurring chip refresh. x.com  x.com  x.com  x.com  x.com  x.com  x.com